Gabriel Lumagui

Always creating.

[Gabriel Lumagui @GabrielLumagui]

@Gstielz one thing i'd watch: the file name is doing a lot of the thinking. "IMG_4821.jpg" is going to get a very confident caption about nothing. passing the image itself to gemini plus a quick approve step before it posts saves you from that

[Gabriel Lumagui @GabrielLumagui]

@GabeJohansson honestly by running one small lead form campaign with your own money. two settings teach you the most: pick "higher intent" on the instant form, and connect it to your email tool so leads get the welcome email right away instead of going cold in a csv

[Gabriel Lumagui @GabrielLumagui]

nobody asked the ai to add product showcase. marketing api v27 may still turn it on if it has "high confidence" a product set matches.

that happens on a new ad with no showcase setting. i'd ask if your tool sends product_extensions OPT_OUT.

[Gabriel Lumagui @GabrielLumagui]

@JakeDSBD24 the first messy input i'd save is the one that already broke it once. every bug becomes a test case, so the next prompt tweak can't quietly bring it back

[Gabriel Lumagui @GabrielLumagui]

@arceyul worth reading meta's custom audience terms before uploading anything like this. you're the one promising meta you have the right to use that data, not the tool

[Gabriel Lumagui @GabrielLumagui]

@RishanX and if it was ads manager, question two is what event it optimized for. a campaign told to find link clicks will find you people who love clicking links

[Gabriel Lumagui @GabrielLumagui]

@MarcusMansory nice, glad it's live. hardest part now is not touching it for a few days. let me know what the first leads look like

[Gabriel Lumagui @GabrielLumagui]

@ZofloqDev both. view source, search for "availability" in the product schema and check it says the same thing as the page. google's rich results test shows what it actually parsed, which settles arguments fast

[Gabriel Lumagui @GabrielLumagui]

google shopping may miss your sale price if javascript adds it after the page loads.

google says googlebot reads your server's html. that price triggers an error.

i'd check a sale item in view page source (not inspect). the price must match your merchant center feed exactly.

[Gabriel Lumagui @GabrielLumagui]

@suvie_writes the tell is when nobody can write the steps down. if you can't list what happens after a lead comes in, the agent will make up its own answer and run it at 2am, with confidence

[Gabriel Lumagui @GabrielLumagui]

@theadgoose the 100-click recheck is a good habit. by then there's enough delivery to tell whether meta's running the version you approved or its own remix

[Gabriel Lumagui @GabrielLumagui]

meta paused the nov 20 terms update, which is nice, except the ai terms have been live since may 2024. if you publish the output, you're "solely responsible" for whether it's accurate. and the feature that rewrites your headlines is already on by default.

[Gabriel Lumagui @GabrielLumagui]

a daily ai report that only says "all good" gets muted by thursday. then one day it isn't.

hermes agent: if the reply contains [SILENT], nothing gets sent. the output is still saved locally.

prompt: if everything's fine, reply only [SILENT]. failed runs still deliver.

[Gabriel Lumagui @GabrielLumagui]

@KevEdelson on #2 i'd fire it on the booking confirmation, not the calendar page load. otherwise meta optimizes for people who like looking at your calendar, which is a different hobby

[Gabriel Lumagui @GabrielLumagui]

@theadgoose the screenshot habit is underrated. the preview is what meta might run, day one is what it actually picked, and those aren't always the same ad

[Gabriel Lumagui @GabrielLumagui]

@morr1ss and on meta the copy can move again after you write it. advantage+ text improvements can pull phrases from your copy and show them "directly or adjusted" as headlines, so i'd check the advantage+ creative tab in advanced preview before launch

[Gabriel Lumagui @GabrielLumagui]

@ZofloqDev stock is the sneakier one because nobody thinks to check it. same test though, if it's not in view source, don't assume google sees it

[Gabriel Lumagui @GabrielLumagui]

@Wv3EcACHtizRjZK yeah, a clean-looking email that shouldn't have gone out still has to count as a fail. the input alone doesn't tell you that.

[Gabriel Lumagui @GabrielLumagui]

an ai workflow fix tested on one message is just crossed fingers.

n8n's evaluation docs: add the input that caused the bug to your dataset, then run the whole dataset over the workflow again before you ship the fix.

that's how you check the fix didn't break something else.

[Gabriel Lumagui @GabrielLumagui]

@BoboLinNotes exactly. every bug you add makes the suite more useful and also more expensive to run, so you feel it when you skip a full pass.

[Gabriel Lumagui @GabrielLumagui]

before your black friday ads go live: ad preview → expand → advanced preview → advantage+ creative tab. use the arrows on each card to see what might actually run. meta says some of these can only be turned off there.

i'd check every card before you publish

[Gabriel Lumagui @GabrielLumagui]

they call it "text improvements," and those same ai terms apply to it. it takes phrases from your copy and shows them "directly or adjusted" as headlines and text overlays.

meaning the claim your page has to back up might not be a line you actually wrote.